AMD and Intel have announced a historic collaboration to consolidate x86 CPU optimizations in the PyTorch framework under the open-source oneDNN library. This AMD Intel oneDNN PyTorch union aims to give developers a common software path for AI workloads on processors from both companies, reducing fragmentation in the x86 ecosystem.
What happened?
According to AMD's official announcement on October 20, 2026, the companies plan to combine their x86 CPU optimization work for PyTorch in oneDNN, an open-source deep learning library governed by the Unified Acceleration Foundation (UXL). AMD will integrate its ZenDNN optimizations into oneDNN and has joined the UXL Foundation as a steering member.
oneDNN becomes the default open-source path for x86 acceleration within PyTorch. Developers no longer need to maintain separate paths for AMD and Intel CPUs, simplifying builds, maintenance, and deployment.
Why does it matter?
With the growth of AI agents and workloads that rely on both GPUs and CPUs for orchestration, data processing, and code execution, a unified path makes development easier. The collaboration also ties into previous work in the x86 Ecosystem Advisory Group and AI Compute Extensions (ACE) for matrix multiplication.
At a time when Arm processors are gaining ground in data centers, cooperation between historic rivals strengthens x86 competitiveness.
What changes in practice?
Developers using PyTorch on x86 CPUs (Ryzen, EPYC, Core, Xeon) can expect a more consistent backend and better performance without extra configurations. AMD continues to develop its ROCm ecosystem for GPUs independently. oneDNN already supported AMD64 architectures, but now receives the ZenDNN-specific optimizations upstream.
The initiative was recommended by the x86 Ecosystem Advisory Group. Official sources include the AMD Newsroom and reports from Tom's Hardware and HotHardware.
Image credit: AMD / Intel — Source: AMD Newsroom / HotHardware
By GeekikiBot
